Which batch API is the fastest?

Right now gpt-5.6-sol at OpenAI has the fastest measured median (79s), against 38 min for the slowest model we can quote. We measure the real queue time, so this is an observation, not a guess.

How much does the batch API cost against synchronous?

50% of the synchronous rate - half price - on both input and output tokens. That is OpenAI, Anthropic and Google's own published batch price, not a batchwatch estimate. batchwatch measures the one thing they do not publish: how long the queue actually takes.

Why do some models show a count instead of a median?

Because a median on a handful of jobs is just the slowest of a handful. We publish a percentile only once a model has at least 8 measurements, so every number in the ranking is one we would stand behind. Until then the row shows the honest count rather than a figure we invented.

batchwatch › Compare

Which batch API is actually fastest right now?

Every batch model we measure, side by side: the median and p90 queue time we observed, the sample size behind each, and how much confidence it earns. Batch costs 50% of the synchronous price at every major provider - this is the wait that comes with the discount, measured.

Right now the fastest is gpt-5.6-sol at OpenAI, with a measured median of 79s — against 38 min for the slowest we can stand behind. That gap is the number nobody else publishes.

The 50% batch price point

50% of synchronous OpenAI, Anthropic & Google's own published batch rate, on every input and output token

The saving is not ours to claim or inflate - it is the providers' own published price, the same at all three: a job that costs $100 synchronously costs $50 in batch. What you trade for it is a queue wait, and that is the number every comparison guesses at and none of them measure. The table below is our measurement of it, one row per model.

The head-to-head

#	Model	Median (p50)	p90	n	Contributors	Confidence
1	gpt-5.6-sol OpenAI	79s	24 min	294	1	Low
2	gemini-2.5-flash-lite Google	2 min	4 min	85	1	Low
3	gemini-2.5-flash Google	2 min	5 min	238	1	Low
4	gemini-3.7-flash Google	2 min	19 min	344	1	Low
5	claude-haiku-4-5 Anthropic	2 min	4 min	354	1	Low
6	gpt-5-nano OpenAI	5 min	6 hours	291	1	Low
7	gpt-5.6-luna OpenAI	38 min	6 hours	302	1	Low

Ranked by measured median, fastest first. Each model links to its full page with the distribution behind the number. Models still building a baseline are listed after the ranked ones, with the honest count.

Questions this answers

Which batch API is the fastest?: Right now gpt-5.6-sol at OpenAI has the fastest measured median (79s), against 38 min for the slowest model we can quote. We measure the real queue time, so this is an observation, not a guess.
How much does the batch API cost against synchronous?: 50% of the synchronous rate - half price - on both input and output tokens. That is OpenAI, Anthropic and Google's own published batch price, not a batchwatch estimate. batchwatch measures the one thing they do not publish: how long the queue actually takes.
Why do some models show a count instead of a median?: Because a median on a handful of jobs is just the slowest of a handful. We publish a percentile only once a model has at least 8 measurements, so every number in the ranking is one we would stand behind. Until then the row shows the honest count rather than a figure we invented.

See it live

Every number here is measured, and it is all public.

The live dashboard shows every model with its full distribution, n and confidence grade. /v1/coverage is the same data as JSON, no key required. Contributing five measurements in seven days cuts the public delay to five minutes and unlocks the decision endpoints (/v1/should-i-batch, /v1/estimate-batchtime).